lip reading through binoculars.
Interesting work, but it still requires a steady handed operator plus stationary and unaware subject, or can they carry out this analysis from composite CCTV footage?
Surely easier, more reliable, but probably less deniable, to intercept at the network message centre end.